S0ulFood/second-opinion ? reverse-engineered prompt

Build me a Claude Code skill called Second Opinion that helps with design thinking by giving an AI four roles, Strategist, Researcher, Executor, and Critic.

I want it to feel like an honest design partner, not a yes machine, so it should question the problem before solving it, check whether assumptions are actually supported, build only once the direction is clear, and then critique the result with a fair severity level and a clear verdict. It should work for different kinds of design work, like product, service, systems, or content, without being tied to one domain.

Please set it up so there is one shared foundation the roles all read first, plus clear role instructions and a simple way to switch between them when the situation changes. Include guidance for when not to trust the skill, and a small validation set or example checks so the behavior can be tested and recalibrated later. If you need current Claude Code skill docs or format details, look them up online first.